tMHG-Finder: Tree-guided Maximal Homologous Group Finder for Bacterial Genomes

Key results

1. ShownProgressive traversal of a guide tree with consensus sequences for each MHG cuts blastn comparisons from O(n^2) to O(n). tMHG-Finder therefore scales linearly with taxa, while MHG-Finder became impractical past a few genomes. Cactus and SibeliaZ remain faster.

   Intel Xeon Gold 5218: MHG-Finder took 54 hours for 9 genomes; tMHG-Finder handled all 30 simulated genomes in 13 hours on a single thread and 7 hours with eight threads; Cactus and SibeliaZ needed 30 minutes, and Mauve 53 hours, for 30 genomes. Authors report an over 10x speedup vs MHG-Finder for fewer than 10 taxa on one thread. · 3.1 tMHG-Finder Runtime Benchmark; 4 Discussion; Figure legends

2. ShownOn a simulated 30-genome ALF benchmark, only tMHG-Finder and Progressive Cactus exceeded an F-score of 0.90. tMHG-Finder had the highest precision and fewer erroneous alignments, but lower recall than Cactus.

   mafComparator against 3989 true MHGs: F-score 0.932 for tMHG-Finder vs 0.966 for Cactus, all others below 0.75; recall 0.877 vs 0.945 (Cactus); incorrect predicted alignments 0.6% vs 1.3% (Cactus); every tool exceeded 97% precision; SibeliaZ coverage was near 50% for nine genomes. · 3.2 tMHG-Finder Accuracy Assessment; 4 Discussion; Tables

3. SuggestedFractionalization is a new asymmetric score, based on overlapping ranges, of how one alignment splits across another alignment set. It showed that SibeliaZ and Cactus over-fragment true MHGs, Mugsy and progressiveMauve under-partition them, and tMHG-Finder output stays closest to the truth.

   Fractionalization runs from 0 (a matching MSA exists) to 1 (heavily fragmented) and was computed in both directions. tMHG-Finder MHGs reach up to 4,500 bp, mirroring the truth, with most values near 0 (some near 0.5). Mugsy and progressiveMauve MHGs extend to roughly 20,000 bp. Most true MHGs score near 1 against SibeliaZ and Cactus. · Fractionalization: MSA Comparison Statistic; 3.2 tMHG-Finder Accuracy Assessment; Figure legends

4. ShownIn 80 Deinococcus-Thermus assemblies, every near-complete 16S rRNA copy fell into one MHG at the root of the guide tree. This supports that tMHG-Finder captures conserved loci, while its heuristics for short sequences keep fragmentary artifacts out.

   Of 312 sequences annotated as 16S (about 50 bp to 1,500 bp), all 162 near-complete copies (1,400 bp) were encapsulated in a single MHG averaging 1,256 bp. The rest, mostly under 200 bp, were judged likely sequencing or assembly artifacts. · 3.3 Validation in Deinococcus-Thermus Genes and…; 4 Discussion

5. SuggestedAcross the internal nodes of the guide tree, more divergent and larger genome sets produced more, shorter MHGs, each covering fewer genes. At the root, conserved single-copy MHGs mostly mapped to single genes such as rpoC, with GO terms that include DNA repair and response to stimulus.

   Root node (80 genomes): 36,282 MHGs, mean 1,008 bp. Bottom-level nodes: under 725 MHGs, max 7,942 bp. Deinococcus clade: 27,066 MHGs vs 8,847 for Thermus/Meiothermus. Genes covered per MHG fell from 4.34 to 0.95. Of 58 conserved MHGs, 38 matched one gene and five spanned several. GO terms came from 54 genes in 44 conserved MHGs. · 3.3 Validation in Deinococcus-Thermus Genes and…; 3.4 Deinococcus-Thermus Evolutionary Rates; Figure legends